Daniel Doyon (@deadly_onion) on X

It's a race to build

of X: • Copilot of Writing (cf

,

,

) • Copilot of Spreadsheets • Copilot of Design As purveyors of fine reading software, we naturally wondered "What about the Copilot of Reading?"

Fortunately, we're in position to explore this question because we're joined by an itinerant

. You might know Simon from some of his viral posts like every.to/superorganizer… by

Simon wanted to hack on LLMs in Reader. We certainly weren't going to say no.

One thing we quickly learned is using generative AI to help you write is obvious; using generative AI to help you read is less so. I won't bore you with all the failed experiments, but we've finally found some Ghostreader use cases with real utility. Let us share some below.

ASK THE DOCUMENT A QUESTION ❓ I'll start with one of the most advanced, 🤯ing examples. I uploaded an esoteric legal document and constructed an prompt pertinent to a commercial real estate loan using our templating language. It nailed it.

Now let's step back. DEFINE 📖 Life-changing? Nah. Lots of reading apps have dictionary lookups. But it is cool to get the right definition for "cords" as in a "cord of firewood" off the bat. I also engineered this prompt to include some spicy emojis 🌶️🤓

ENCYCLOPEDIA LOOKUP 📕 This is the encyclopedia use case. This prompt works best when dealing with the kind of person, place, or concept you'd be able to find in an external encyclopedia such as Wikipedia. For internally defined terms, see COMING TO TERMS below.

TRANSLATE 🌎 GPT-3 is pretty good at translating from one language to another, but your mileage will vary based on the particular language. GPT can even translate to dialects. For laughs, try translating a passage to Jar Jar Binks.

COME TO TERMS 🤝 Now we're getting into innovative territory. If you study the art of reading, it won't be long before you come across the concept of "coming to terms". Words have many meanings and it's your job as reader to "come to terms" with the writer's use of language.

Matter of fact, you and I just came to terms on the meaning of come to terms. You can use Ghostreader to "come to terms" faster. If you've read Seeing Like a State, you'll note that GPT nailed a *simple* explanation of its key term legibility (cf

ribbonfarm.com/2010/07/26/a-b…)

SIMPLIFY 👶 Simplify is my favorite pre-engineered prompts. It rewrites complicated text into simpler terms. This won't do much for low density content like news articles, blog posts, Twitter threads, but it's amazing for research papers, internal memos, and challenging books.

SUMMARIZE💧 Anyone entering the Copilot of Reading idea maze will be first drawn to summarization. After all, if some creator is tapping +++ in his word processor to spew forth the next few paragraphs, shouldn't we readers be able be able to tap --- to do the opposite?

For reasons I'll skip here, summarization as a substitute for reading is (generally) lame. That said, summarization as an *aid* to reading can be cool. Our favorite use of summarization is at the section level. Ghostreader can operate on all text contained in a section.

If you must, you can also summarize whole documents. For the GPT nerds reading this thread, the GIF below shows me using one of the truncating functions built into our templating language to reduce a long document to something GPT can handle.

WHIMSICAL PROMPTS 🙃 We've included a handful of silly prompts with Ghostreader not because they're particularly useful, but to help folks learn how to ride this bicycle for the mind that is generative AI.

HAIKU ☮️ For some odd reason, this prompt often achieves levels of compression exceeding more explicit prompts such as "summarize" or "tl;dr". I suppose it has something to do with the strict constraints learned by GPT-3.

EMOJIFY 🚴🧠 Turn selected passage into a series of emojis. Emojis actually do have utility when sharing annotated docs because it makes your notes far more delightful for your friends, family, and colleagues to read. More on sharing: x.com/ReadwiseReader…

CUSTOM PROMPTS ✍️ These were all just built-in pre-engineered prompts. Perhaps more importantly, you can write your own custom prompts using our templating. This lets you access variables like the document content, title, highlights, and more to construct powerful prompts.

This is *just* the v1 alpha to get the feedback loop going. We're already deep into v2 which weaves Ghostreader into the mobile apps and browser extensions among many other exciting innovations.

One innovation is a powerful new primitive: CHAINING ⛓️ Chaining enables you to carry one GPT response into your next prompt yielding far better results on complex instructions. I'll bet this UX pattern becomes one of the next big things in application layer generative AI.

is still in private beta but we're rolling folks off the waitlist and will be public soon. If you want in now, DM me or

the email you use with

and we'll hook it up (might take a few days as we're in BA for an offsite). readwise.io/read

Why the cheeky name "Ghostreader"? 👻🤓

came up with and I instantly loved it I grew up watching a pseudo-educational show on PBS called Ghostwriter in which a band of city kids solve crimes with the help of a mysterious intelligence locked away in a desktop computer.